title: bn-play Presentation output: ./dist/presentation.html controls: true
--
Josh Gribbon
--
- Motivations
- Goals
- Outcomes
- Process
- Future work
--
- Value of statistical and causal knowledge
- Ease of access of the web
- Simple and fast interface
- Short feedback loop
--
- Philosophical basis - causality is important to know how to intervene
- People not understanding statistics seems to lead to a lot of drama and confusion
- Can affect regular daily decisions, legislature, etc
--
- Graphical model for conditional dependence
- Directed Acyclic Graph (DAG)
- Terminology
- For causality, mainly:
- CMC - node independent of non-descendants given parents
- Faithfulness - structure maps well to data and captures relationships
- Causal sufficiency - all common causes measured
- VS completeness w/ all in the graph
--
- Build and change the BN visually
- View properties of the nodes - CPT, relatives
- Edit CPT values
- Run simulations, with interventions
- View results
--
- Super important!
- Clear information
- Coordinated data
- Adaptive layout
- Useful visualizations
- Prompting and guiding interaction
--
* Seemed simple * Looked useful for DBN in the future * Intermediate language got complicated--
* More flexible API * Better options for expansion * Just the right amount of generality--
--
--
Powerful and worth mentioning
Vue.js
- frontend components, templating, data binding, build systemBootstrap
,Webpack
,Browserify
Vis.js
,Chart.js
- visualizationsFirebase
- hosting, free until ~14k initial downloads per month
--
<iframe src="https://bn-play.firebaseapp.com/" width="100%" height="85%"></iframe>--
- Simple site to create BNs
- Change the structure
- Change the CPT for a node
do()
on a node- Run simulations
- Visual and textual information and guidance
--
- Teaches aspects of causality, doesn't identify it
- Con: Isolated from the data
- Pro: Avoids many concerns
- Doesn't identify cycles or explain the issues with them
- Similarly, issues like cancelling out that violate faithfullness
- Can't detect structure from data - useful but complicated to automate
--
- Get feedback for new functionality - most useful thing
- "Realtime" simulation
- Import/export network or simulations
- Detect cycles in graph
- List some conditional independencies and the methods for identifying them
--
- Showing d-seperation, d-connection
- Toggle hierarchical layout
- Edit node details - name, color, states
- Learn network structure
- Deploy as native app - connectivity benefits
--
https://bn-play.firebaseapp.com/